SURFACE EMG SIGNAL GESTURE RECOGNITION BASED ON HYBRID DILATED CONVOLUTION CNN AND BIGRU

Aimed at the problem of low accuracy and large amount of calculation for gesture recognition based on surface electromyography (sEMG), a method for sEMG gesture recognition based on a hybrid dilated convolutional neural network combining bidirectional gated recurrent unit and attention mechanism is proposed. Compared with the ordinary CNN, HDC can expand the receptive field, reduce over-fitting, and extract more features by setting the dilation rate to parity hybrid and different sizes. The BiGRU module can extract and process the timing features of the data well, and attention module gives greater weight to important features, which can improve accuracy. The accuracy rates of 92.72% and 97.85% were achieved on the NinaproDB1 dataset and the self-acquisition dataset, respectively.